It is well understood in the art of data processing that it is desirable to interconnect local networks. In the prior art such local networks have been coupled together through devices generally referred to as routers. Such router devices are normally peculiar to the manufacturer thereof, such that in order for one station, on a local network, to send messages to a second station, on another local network, the first station must transmit at least a two level address or involve a two level protocol arrangement. In other words the address arrangement in the prior art is analogous to a common telephone system wherein the first three digits, i.e. exchange number such as Gramacy 3 (GR3), is one level of address and the last four digits is a second level of address. The foregoing arrangement has been reasonably acceptable. However as users have commenced purchasing equipment from different manufacturers, problems have arisen with respect to communicating, from one network to another, through a router which is manufactured by a different manufacturer than the manufacturer of the stations communicating. The difficulties have arisen in part because the higher levels of protocol required in prior art systems and also because a station of a different manufacturer has great difficulty in providing information to a router of a first manufacturer even if the router has memory means. The present invention has only one level of address information, has a learning capability to increase its knowledge as to what stations are where, on any network, connected thereto, and includes a means for jettisoning relatively unused address information.